Image Denoising Using a Modified LNMF Algorithm

A novel image denoising method is proposed in this paper by using local non-negative matrix factorization (LNMF) with sparse constraint, denoted by SC-LNMF. LNMF method can successfully extract the local feature of a nature image and denoise images efficiently. However, LNMF method does not consider the image´s sparse prior distribution and the sparse control of feature basis vectors and sparse coefficients. To enhance the feature matrix´s sparseness and the feature sub-space´s locality, SC-LNMF is proposed here. Using 10 clear images to learn the SC-LNMF algorithm, simulation results show that this method is indeed efficient in extracting images´ local features. Further, considering different noise variance, utilizing the SC-LNMF feature bases, the noise was reduced hardly. At the same time, using the signal noise ratio (SNR) measure to evaluate denoised images´ quality, simulation results testify that our method proposed here is effective and feasible in performing image denoising task.